Scientists Created New AI Tool to Diagnose Parkinson’s and Heart Failures Using Eye Images

Scientists Created New AI Tool to Diagnose Parkinson's and Heart Failures Using Eye Images

Scientists have introduced an artificial intelligence (AI) tool named RETFound. This new AI tool is capable of diagnosing and predicting various health conditions based on retinal images.

While AI tools have previously been used for disease detection in retinal images, RETFound distinguishes itself through its application of self-supervised learning. This approach eliminates the need for manually analyzing and labeling the 1.6 million retinal images used in training. Those typical machine-learning models are both time-consuming and expensive.
RETFound employs numerous retinal photos to predict the appearance of missing portions within images. It’s like the training of large language models like ChatGPT.

“Over the course of millions of images, the model somehow learns what a retina looks like and what all the features of a retina are.”

Pearse Keane, an ophthalmologist at Moorfields Eye Hospital NHS Foundation Trust in London and co-author of a paper in Nature

Why retinal images?

Retinal images provide valuable insights into a person’s health since they are the only body part that directly reveals the capillary network, the smallest blood vessels.

“If you have some systemic cardiovascular disease, like hypertension, which is affecting potentially every blood vessel in your body, we can directly visualize [that] in retinal images,”

Pearse Keane, an ophthalmologist at Moorfields Eye Hospital NHS Foundation Trust in London and co-author of a paper in Nature

Moreover, retinas share similarities with the central nervous system, making them suitable for evaluating neural tissue.
Researchers initial pre-training on the 1.6 million unlabeled retinal images. And also introduced a small number of labeled images representing specific conditions, such as Parkinson’s disease. Leveraging its understanding of normal retinal appearances acquired from the unlabeled data, the model quickly learned to identify retinal features associated with diseases.

High-quality labels for medical data are extremely expensive, so label efficiency has become the coin of the realm

Curtis Langlotz, director of the Center for Artificial Intelligence in Medicine and Imaging at Stanford University,

RETFound demonstrated impressive performance in detecting ocular diseases. Particularly diabetic retinopathy, scoring between 0.822 and 0.943 on a scale where 0.5 represents random prediction and 1 illustrates perfect accuracy. When predicting systemic diseases like heart attacks, heart failure, stroke, and Parkinson’s, its overall performance, while limited, surpassed that of other AI models.

Expanding to new areas

RETFound stands out as one of the few successful applications of a foundation model to medical imaging. Researchers are now exploring applying similar techniques to more complex medical images, such as magnetic resonance or computed tomography scans.
The model is publicly available, with hopes that research groups worldwide can adapt and train it for their specific patient populations and medical contexts. However, caution is needed, as any limitations within RETFound could carry over into future models built upon it. Transparency and ethical usage are paramount as this technology advances in the medical field.

Also read;

Another Death Records after Brain Eating Amoeba Infection at Little Rock Country Club